Voicing your words : Headset that gives voice to the voice-boxless

Things that we all take for granted, even a mundane thing like talking becomes incredibly painful if you have to face a situation where no words will come out of your mouth when you are actually mouthing them. This is the distressing and frustrating situation faced by people who have had their larynx removed following cancer, disease or injury. To help them communicate, they are often fitted with a valve in their throat to divert air from the lungs to the esophagus when they exhale, generating a form of speech. But these valves tend to become clogged after only a few months and need to be replaced.

voice box less headset

In order to remedy this situation, a team from the UK is developing a device that can detect and interpret facial movements when someone mouths a word, recognizing what they are saying. The device uses small magnets placed inside the mouth and on the tongue to create a magnetic field. Changes in the field as the person mouths a word are detected by sensors attached to an external headset. The system is trained to recognize the patterns of positional changes that correspond to the individual wearer mouthing particular words.

So far, the system can recognize about 50 words. The team plans to develop magnets that can be safely implanted into the tongue. The researchers are also aiming to reduce the size of the headset that holds the sensors down to that of a Bluetooth device. Still, it will be a challenge to arrange the magnets in such a way that they produce enough information to recognize what the wearer is saying without causing them discomfort.

While losing your speech entirely is thankfully relatively rare, thousands of people each year who suffer a stroke - or those with cerebral palsy, Parkinson's disease or motor neuron disease - can lose the ability to speak coherently. Just as human listeners have trouble understanding impaired speech, so do conventional speech-recognition systems, say the experts.

This system is trained on a number of recordings of the person's voice so that it learns to recognize their individual speech patterns. Once the system has recognized a word, it is replayed by a voice synthesizer. Alternatively, a few minutes of voice recording is taken from a family member - or the person themselves, if made before their impairment becomes too severe - and used to adapt a standard artificial voice. Devices that can clarify impaired speech are far quicker than relying on typed words.

The team's prototype was developed in collaboration with the the National Health Service, communication device maker Toby Churchill and speech-recognition specialist El Pedium Technologies, all in the UK. Patient trials are supposed to be starting within the next 12 months.

Via: New Scientist